QoS-aware mobile service transactions in a wireless environment

CONCURRENCY AND COMPUTATION: PRACTICE AND EXPERIENCEConcurrency Computat.: Pract. Exper. 2007; 19:1219–1236Published online 3 April 2007 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/cpe.1157

QoS-aware mobile servicetransactions in a wirelessenvironment

Muhammad Younas1,∗,†, Kuo-Ming Chao2,3,Ping Wang4 and Chun-Lung Huang5

1Department of Computing, Oxford Brookes University, Oxford OX33 1HX, U.K.2Software School, Fudan University, China3Department of Computer and Network Systems, Coventry University, Coventry CV1 5FB, U.K.4Department of MIS, Kun Shan University of Technology, Taiwan, ROC5Institute of Information Management, National Chiao Tung University, Taiwan, ROC

SUMMARY

Mobile computing greatly facilitates the provision of mobile services using portable and handheld deviceswhich are connected through wireless networks. A pre-requisite for the provision of an effective service is toensure quality of service (QoS) during the concurrent execution of requests and system or network failures.This paper presents a new transaction management approach for the provision of mobile services whichis based on advanced transaction models and fuzzy selection criteria. It employs a novel QoS consensusmoderation approach in order to select preferred mobile services from among various alternatives, thusproviding multiple pathways to the completion of mobile service transactions. Unlike traditional transactionmanagement schemes, the proposed approach takes into account QoS attributes in the commit process ofmobile service transactions. The proposed approach is implemented as a prototype tool and evaluatedthrough a number of experiments. Experimental results show that the proposed approach effectivelyselects mobile services and significantly increases the throughput (i.e. commit rate) of mobile servicetransactions. Copyright c© 2007 John Wiley & Sons, Ltd.

Received 30 October 2006; Accepted 21 November 2006

KEY WORDS: mobile services; mobile transactions; Web services; quality of service; fuzzy logic

1. INTRODUCTION

Currently, an increasing number of people are using mobile devices (such as mobile phones,PDAs, etc.) in order to access information services in a timely and flexible manner.

∗Correspondence to: M. Younas, Department of Computing, Oxford Brookes University, Oxford OX33 1HX, U.K.‡E-mail: [email protected]

Contract/grant sponsor: National Science Council of the Republic of China; contract/grant number: NSC 95-2416-H-168-007

Copyright c© 2007 John Wiley & Sons, Ltd.

1220 M. YOUNAS ET AL.

Examples include obtaining stocks quotes, weather information, and news, and conducting financialtransactions. The utilization of such information services is further facilitated through the Web servicestechnology [1], which enables software applications to seamlessly interact and integrate using standardtechnologies such as XML, SOAP, WSDL, and UDDI. In the following discussion we use the termmobile services (as in [2,3]) to refer to Web services in a wireless environment. That is, Web serviceswhich are accessed through mobile devices are termed as mobile services. For instance, a hotel bookingservice (exposed as a Web service) can be accessed through a mobile device in order to make a hotelbooking. Furthermore, such services can be composed with other services (such as flight booking)which can be collectively used by mobile users in order to acquire multiple services with a singlerequest.

Although mobile computing greatly facilitates the provision of services it is associated withvarious QoS issues such as performance, throughput, reliability, consistency, security, and usabilityissues. In this paper we investigate the transaction management of mobile services, as this iscrucial to maintaining the correct execution of services and the consistency of their underlying data.Transactions have been employed in various applications, such as e-business and mobile commerceapplications [4,5]. The classical definition of a transaction is that it comprises a sequence of actionssuch that either all of them are completed or none. Classical transactions are based on the atomicity,consistency, isolation, durability (ACID) properties [4]. Atomicity enforces an ‘all or nothing’ policysuch that a transaction must either appear to execute in its entirety or not at all. Consistency demandsthat a transaction must transform data from one consistent state to another consistent state. Isolationrequires that an active transaction cannot expose its intermediate results to other transactions beforeits commitment. Durability means that the effects of the committed transactions must not be erasedand should be made permanent in persistent storage for recovery purposes. In order to relax the ACIDproperties, various advanced (or extended) transaction models have been proposed in the literature,including open-nested transactions, flexible transactions, conversational transactions, and so on [4,6].These models mainly relax atomicity and isolation properties. For example, atomicity is relaxed byallowing the partial commitment of a transaction and isolation is relaxed (or violated) by allowingtransactions to see the intermediate results of other transactions.

Appropriate transaction management is important, specifically from the mobile user’s perspective,as it should guarantee that services obtained are consistent with what users believe they have requestedand what the system has delivered. From the service provider’s perspective, it must ensure thattransactions are correctly executed and that the information held in service provider data sources isconsistently maintained. Existing approaches are limited to classical QoS such as response time, systemand network efficiency, and the reliability of mobile transactions. They do not consider QoS attributesspecific to mobile services such as execution price, usability, and so on [7]. In this paper we proposea new transaction management approach that takes into account classical as well as mobile service-specific QoS attributes.

Our approach is based on advanced transaction models and the QoS consensus moderationapproach (QCMA) [7]. As described above, advanced transaction models relax ACID properties.Such models allow for the partial commitment of mobile transactions in contrast to ACID-based‘all or nothing’ commitment. Thus they are more appropriate for mobile service transactions(MSTs) owing to the nature of wireless computing environments wherein long delays, failures, anddisconnections are frequent. Traditional ACID properties may result in long delays and in uselessconsumption of resources (e.g. locking data). Furthermore, advanced transaction models allow for the

Copyright c© 2007 John Wiley & Sons, Ltd. Concurrency Computat.: Pract. Exper. 2007; 19:1219–1236DOI: 10.1002/cpe

QOS-AWARE MOBILE SERVICE TRANSACTIONS IN A WIRELESS ENVIRONMENT 1221

execution of alternative transactions. That is, if a component transaction of a MST fails to acquirea requested service, then an alternative will be executed in order to acquire an alternative service(if it exists). The use of alternative transactions is useful for two reasons: first, it provides userswith a choice of various services; second, it increases the transaction throughput (i.e. commit rate)by preventing transaction failures. However, the selection of appropriate alternative services (foralternative transactions) is complicated by the availability of a large number of alternative services.For instance, a hotel booking service can have many alternatives provided by different serviceproviders. Various elements contribute to the complexity of the selection of appropriate services:(i) the large number of alternative services provided by different service providers; (ii) the various QoSattributes such as execution price, efficiency, usability, security, and availability; (iii) the differences inthe preferences and opinions of users as well as service providers. The proposed QCMA deals withthese complexities and provides a method for selecting appropriate services. It assigns different QoSweights to services and ranks them accordingly. A MST can then use appropriate services in order tofulfill a user’s requests. That is, it uses the highly ranked service first to perform the user request. If thisservice fails, then it uses the next highly ranked service, and so on. The aim of QCMA-based selectionis twofold: first, it ensures the provision of a QoS-based service; second, it facilitates the selection ofalternative services that are used to increase the throughout of a MST.

The proposed approach is based on our previous work of mobile transactions and Web services [7–9].To the best of our knowledge, the existing research does not consider QoS aspects in mobile (service)transactions.

The remainder of this paper is organized as follows. Section 2 reviews current work and identifiesrelated issues and Section 3 describes the basic concepts of fuzzy set theory. Section 4 presents theproposed approach and Section 5 presents the evaluation of the proposed approach. Section 6 concludesthe paper.

2. RELATED WORK

MST management is complicated mainly due to the characteristics of mobile devices and wirelessnetworks [8,10]. For instance, the mobility of processing units (e.g. laptop or PDA) means MSTsand data stores may physically re-locate during execution. A MST may originate at one site andterminate at another, and may require frequent re-connections in between. For example, a MST issuedat cell 2 (C2) may terminate at cell 1 (C1) in a wireless cellular network as its originating devicemoves (see Figure 1). Furthermore, the bandwidth of wireless networks is currently low in comparisonwith wired networks. Thus, a wireless network may be overloaded by a high volume of informationexchange. In addition, mobile devices are currently less reliable and have fewer resources than theconventional ‘stationary’ units. For example, the limited power supplies of mobile devices and therelatively limited resources affect throughput of mobile services, because they increase the probabilityof transaction failure (e.g. owing to a flat battery or running out of disk space).

In order to address issues that arise as a consequence of the characteristics of mobile devices andwireless networks, various approaches have been proposed in the literature. Alvarado et al. [11] presenta self-adaptive component-based approach for mobile transactions. This approach does not depend ona single commit protocol. Instead, it implements different protocols (such as two-phase commit (2PC),presumed abort (PA), and presumed commit (PC)) in order to choose an appropriate protocol that suits



Laptop

C1

C4 C2

C3

C5

PTSN MSC

Figure 1. Generalized architecture of mobile computing.

the environment or context of a mobile transaction. For example, the 2PC and PA protocols can be usedif the abort rate of mobile transactions is higher. Performance experiments show that the self-adaptiveapproach performs better than a (pre-defined) single protocol. Varshney and Ravikumar [12] propose atransaction and network-based approach for group-oriented mobile services. This approach identifiestransaction requirements for such services and accordingly develops network protocols. It presentsvarious simulation experiments in order to evaluate different types of transactions. The proposedprotocols are shown to provide transactions with higher throughput (i.e. higher completion probability).

Kumar et al. [10] define a protocol, called TCOT, for mobile transactions. TCOT, a variant of theclassical 2PC protocol, is based on a timeout mechanism and is claimed to have improved performanceand throughput over the 2PC protocol in managing mobile transactions (the 2PC protocol enforcesclassical ACID criteria). However, 2PC and ACID criteria are inappropriate for MSTs. Lee andHelal [13] introduce a mobile transaction model, called high commit mobile transactions (HiCoMo), inorder to improve the commitment rate of mobile transactions. Using HiCoMo transactions, aggregateddata can be updated in situations where mobile units are disconnected.

The Kangaroo Transaction (KT) model [14] takes into account the movement behavior of mobiletransactions so that they can hop from one base station to another as their mobile device moves.Such transaction hopping is modeled through the use of a split transaction model. In the KT model,a transaction splits only when it hops from one base station to another. Splitting is not allowed iftransaction remains in the same cell. The KT model aims to reduce message overhead in the followingway. When a mobile transaction moves to a new cell, the control of the transaction also moves to thenew cell. However, if it were to remain at the originating cell messages would have to be sent from theoriginating cell to the current base station whenever the mobile unit requests information. To avoid thismessage overhead, the transaction management function should move with the mobile unit. However,if the originating device of a transaction moves back to the previous base station (or cell), then theinformation related to the transaction management needs to be sent back to the previous station. Thus,in certain situations, it may not help in reducing communication overhead.



A pre-serialization technique for managing transactions in a mobile multidatabase environmentis proposed in [15]. This approach aims to maintain the isolation property of mobile transactions.The benefits claimed include supporting disconnection and maintaining transaction isolation. However,the main limitation of this approach is that these benefits are gained at the cost of processing overhead.The approach presented in [16] proposes a pre-write operation technique. The main objective of thisapproach is to increase data availability and reduce transaction processing on mobile devices. In thistechnique, a transaction is executed on the mobile device using operations such as pre-writes and pre-commit. The pre-write operation is performed on the local data items locked by the mobile device,and pre-commit is used to locally commit the transaction at the mobile device. After local commit,pre-writes of the data items are sent to the database, which makes them permanent and commits themobile transaction.

Dunham and Kumar [17] study the impact of mobility on transaction management (IMTM).The IMTM study analyzes the performance of mobile transactions according to the mobility of differenttransaction management schemes. That is, the transaction management can: (i) be fixed on the mobileunit home site; (ii) be fixed on the anchor site of mobile unit; and (iii) move with the movementof mobile unit. This study shows that none of these management schemes always achieves a betterperformance. Younas et al. [8] proposes a pre-emptive resume scheduling mechanism that exploits theunderlying network capabilities in order to improve the performance of mobile transactions.

The above approaches do not consider the provision of alternative services nor do they give attentionto QoS-based service selection in mobile transaction management. Menasce [18] studies the QoSof component Web services in terms of cost and execution time. This study employs probabilitytechniques to measure the cost and execution time of component Web services by considering differentexecution scenarios such as parallel, sequential, fastest-predecessor-triggered, and so on. This studyhelps in selecting appropriate component services for Web service composition. However, this researchdoes not focus on MSTs.

3. PRELIMINARIES

This section illustrates the fundamental aspects of fuzzy set theory on which the proposed QCMA [7,9]is based.

Ordinary or crisp set theory enforces an all or nothing membership policy. Fuzzy set theory relaxesthis policy and allows the inclusion of fuzzy (or vague) terms. These terms are associated withmembership values which range between 0 and 1 (see [19]). Fuzzy set theory allows for partialmembership wherein 0 represents non-membership and 1 represents complete membership of a term.

In many situations the classical set theory is inappropriate as it precludes the modeling of(alternative) services with marginal values. For instance, a service with a 20 ms response time can beconsidered as an efficient service (i.e. membership value = 1). However, a service with a slightly lowerresponse time, say 19 ms, cannot be considered as efficient (i.e. membership value = 0). Thus, thelatter cannot be considered as an alternative to the former. Fuzzy theory can be usefully applied tosuch a situation. A service with 19 ms response time can be represented as a nearly efficient service(i.e. membership value = 0.9), which can be considered as an alternative service in order to fulfill userrequirements.

A fuzzy set can be defined in two ways: (i) by calculating membership values of those membersin the set separately; or (ii) by defining membership function mathematically. In general, the former



1

ac d+bdc-b

A~ )(a

Figure 2. Trapezoidal membership function [6].

is used when the set is composed of discrete members and the latter is used when the domain isa continuous variable. For example, a fuzzy set (A) can be defined through enumeration using thefollowing expression [19].

A =∑

µA(xi)/xi = µA(x1)/x1 + µA(x2)/x2 + µA(x3)/x3 + µA(x...)/x...

where the summation and addition operators refer to the union operation and the notation µA(xi)/xi

refers to an element xi with a membership degree µA(xi). Those elements xi whose membership valueis zero are not represented.

A fuzzy set for continuous members is shown as follows (four common types for the membershipfunction are trapezoidal, Gaussian, and S- and Z-membership functions):

A =∫

x

µA(x)/x

Using fuzzy set theory, a mobile user’s opinion can be represented in a trapezoidal membershipfunction that can be specified by four parameters:

µA(a) =

x − (c − b)

b, c − b ≤ a ≤ c

1 c ≤ a ≤ d

−x + (d + b)

b, d ≤ a ≤ d + b

0, a > d + b, a < c − b

For example, Figure 2 shows a trapezoidal membership function for a fuzzy set (A) with parameters(c − b, c, d, d + b).



4. THE PROPOSED APPROACH

This section presents the proposed approach which comprises three parts. Section 4.1 describes thetransaction model of mobile services. Section 4.2 presents the QCMA method for selecting appropriatemobile services. Section 4.3 illustrates the execution protocol of MSTs.

4.1. MST model

The model comprises a set of mobile service providers that provide various services (e.g. hotel booking,stock trading, weather forecast, and banking). Each of the services is associated with underlying datarepresenting information such as the type and price of a hotel room. A mobile service user may requesta number of such services using a mobile device.

A MST is a sequence of different actions, where each action represents a component mobiletransaction (CMT), which is executed in order to perform a requested task (or acquire a requestedservice). The sequence is terminated with commit or abort, where commit marks the successfulexecution and abort marks the unsuccessful execution of a MST. Each CMT has a type and comprisesa sequence of operations, which concludes with either abort or commit. A CMT is compensatableif its effects can be semantically undone by executing a compensating service transaction. It is non-compensatable if its effects cannot be undone by executing a compensating service transaction. A CMTis replaceable if there is an associated alternative service transaction. A CMT is non-replaceable if thereis no alternative service transaction associated with it. A CMT is vital if its abort results in the abortof a MST. It is non-vital if its abort does not result in the abort of MST. For example, a CMT of hotelbooking can be considered as vital, while a CMT of car rental can be considered as non-vital if the userspecifies the hotel booking as mandatory and car rental as optional services.

In the proposed approach, a MST is characterized by the following properties that are based onadvanced transaction models [4,6,20].

Relaxed atomicity

This is weaker (more relaxed) than the classical atomicity. Whereas classical atomicity requires aMST to complete successfully or not at all, relaxed atomicity allows the partial commitment of suchtransaction. That is, a MST can still be committed if some of its CMTs are committed. This typeof atomicity also allows the unilateral commit of CMTs irrespective of the commitment of theirsibling CMT, with the constraint that information sources must remain consistent after the executionof such a transaction. Such unilateral commitment is useful for mobile services owing to their complexnature and longer execution times. Relaxed atomicity also allows for executing alternative CMTs.As described earlier, there exist various alternative services that can be executed in case of failures.For instance, if a room cannot be booked in one hotel then it can be booked in an alterative hotelprovided that the alternative room is acceptable to the user.

Consistency

By enforcing relaxed atomicity intermediate states of CMTs become accessible. Thus, consistency canbe enforced only at the component service level. For instance, if a CMT books a hotel then it will



release the resources (e.g. locks) of the hotel data source so that other transactions can make flightbookings. The traditional notion of consistency and isolation of ACID criteria [4] cannot be enforcedin mobile services as they either result in the useless consumption of resources or resource blocking.That is, they require all of the CMTs to wait for the completion of each other in the prepare-to-commitstate.

Durability

As in the ACID criteria, this requires that effects of a committed MST must be made permanent in therespective data sources in order to ensure failure recovery.

4.2. Mobile service selection

The process of selecting appropriate mobile services for a MST is a non-trivial task. As describedearlier, it involves various factors including the availability of a large number of alternative services,the choice of QoS attributes, and the distinct preferences and opinions of users. For instance, a servicewith a 20 ms response time and 100 pence execution price can be considered as efficient and cheap bysome users while others may consider this service as inefficient and expensive. It is therefore importantto have a classification of alternative services for selection according to their overall quality rating.A ranked list of alternative services could facilitate users to effectively select alternatives services.The overall quality rating (or QoS) of a specific service is a view collected from a group of mobileusers who use the service. In order to deal with the differences in the opinions and preferences ofmobile users, we employ QCMA [7,9,21] which enables a group of users to reach a consensus on theQoS rating of a specific service. Such group consensus on the QoS rating is a basis to determine serviceranking in the selection process.

Recall that there exist various QoS attributes for mobile services such as price, reliability,performance, integrity, adaptability, and bandwidth. It is therefore difficult to achieve a consistent viewof a service QoS among different mobile users. Moreover, it is inappropriate to set a strict thresholdto classify alternative services. Thus, in the proposed approach, the opinions of mobile users arerepresented via fuzzy logic. In the following we describe the service ranking process using QCMA(see [7,9] for further details on QCMA). QCMA, based on the fuzzy set theory [22,23], takes intoaccount mobile users’ opinions on a set of QoS attributes in order to achieve an overall quality ratingfor a specific service.

QCMA comprises two sub-processes: the similarity aggregation method (SAM) and the resolutionmethod for group decision problems (RMGDP). The group consensus on a specific QoS attributeis gathered by SAM sub-process and the weightings over different QoS attributes are conducted byRMGDP sub-process (see [9,24,25] for details on SAM and RMGDP sub-processes). SAM aggregatesdifferent mobile users’ fuzzy opinions to form a group’s fuzzy consensus opinion on a specific QoSattribute. Each user represents their subjective fuzzy preference on one specific QoS attribute with apositive trapezoidal fuzzy number. SAM employs a similarity measure to calculate the differencesbetween individuals within the group in order to obtain an index of consensus. The indexes ofconsensus for all pairs of individuals can be used to form an agreed group fuzzy opinion. SAM ensuresthe consistency of the definitions of fuzzy terms for providers and consumers [19].



iBookMH MH

MSC

FH FH

FH

Figure 3. Underlying architecture of the proposed system.

The objective of RMGDP is to resolve group differences and to reach a group consensus on theweightings over different QoS attributes. This sub-process can be divided into three phases: (1) thetransformation phase—transform the individuals’ opinions on different QoS attributes into preferencevalues; (2) the aggregation phase—aggregate the individual preference values for obtaining the grouppreference using an ordered weighted averaging (OWA) operator [26]; and (3) the exploitation phase—compute the weightings of QoS attributes by group preference [19].

The group consensus on QoS attributes are updated periodically by the QoS fuzzy moderator inline with the interactions taking place among mobile users. The QoS and their moderated opinionsare represented in OWL [27] as the ontology. After the QCMA process, mobile users’ opinions ona set of QoS attributes will be aggregated into a group consensus of opinion, which is representedin the form of a fuzzy trapezoidal membership function and is used to evaluate the overall qualityrating for each alternative service. That is, each alternative service will be rated by a fuzzy value.According to the fuzzy values, alternative services can be ranked and listed in ascending order for theselection of alternative services. An alternative service with higher rank indicates a higher possibilityfor substitution which also improves the throughput in the case of transaction failures.

4.3. The execution protocol

The protocol is implemented as one coordinator and several component service coordinators (CSCs).The coordinator is deployed at a fixed host (FH) where CSCs can be deployed at fixed as well as mobilehosts (MHs) (as shown Figure 3). The selection process is deployed at the FH. The coordinator andeach CSC maintain log files in order to record the required information (such as start, commit, andabort) about the execution of a MST and its component MST.

The first step of the protocol is to discover and select the required mobile services (using the selectionprocess). When the required component services are identified, the system starts executing the MST.Each MST is associated with one coordinator and several CSCs that coordinate the execution of theCMTs. The coordinator and CSCs communicate with each other during the execution of the MST.



select service

Start

Recording

send abort to CSC

Committed Aborted

check received-votes

Checking

select alternative

Selecting

check cmt type

Checking

record cmt abort

Recording

record cmt commit

Recording

check votes

Checking

receive votes

Wait

contact CSCs

Contact

[vote = commit]

[vote = abort]

[cmt=replaceable]

[cmt replaceable

cmt vital]

[cmt replaceable

cmt=vital]

[received-

votes=desired-num-

commit]

[received-

votes desired-num-

commit]

Figure 4. Coordinator state diagram.

We will describe two algorithms that illustrate the processing of our protocol. For the sake of simplicity,the coordinator is shown to communicate with one CSC. Algorithms 1 and 2 respectively describe theworking mechanisms of the coordinator and a CSC. These algorithms are diagrammatically representedin Figures 4 and 5, respectively, using UML state diagrams.



record compensation

Compensating

Committed Aborted

check final decision

Checking

receive final decision

Waiting

send cmt abort

Sending

send commit vote

Sending

check cmt

Checking

process cmt

Processing

record cmt start

Start

[cmt = commit]

[cmt = abort]

[final-

decision=commit]

[final-decision=abort]

Figure 5. State diagram for component service coordinator.



Algorithm 1: Coordinator

1. receive request for a new MST2. record start(MST) in a log-file3. select appropriate service //using selection process4. contact CSCi regarding CMTi ∈ MST5. wait to receive votes from all CSCs6. receive votei from CSCi

7. check the status of votes received:if votei = commit

record commit(CMTi) in log fileelse if votei = abort {

record abort(CMTi) in log fileif CMTi = replaceable {

check set of alternative CMTrepeat steps 3–7

}else if CMTi �= replaceable ∧ CMTi �= vital

check received-votes to decide on MST abortelse if CMTi �= replaceable ∧ CMTi = vital {

record abort(MST) in log filesend abort to all CSCs

}}if received-votes = desired-num-commit{

record commit(MST) in log filesend final-commit to all CSCs

else if received-votes �= desired-num-commit{record abort (MST) in log filesend final-abort to all CSCs

}8. terminate MST

Algorithm 1. A new MST is assigned to a coordinator (step 1). It records the start of the MST in a log file(step 2), selects appropriate (highly ranked) service (step 3), and contacts the CSC to initiate CMTs (step 4).The coordinator then enters into a wait state, awaiting messages from CSCs concerning completion of the CMTs(step 5). The coordinator receives votes from the CSCs (step 6) and checks whether these votes represent a commitor an abort of the CMTs (step 7). If a vote received is a commit, it records this in a log file and proceeds to thenext. If a vote received is an abort then it checks whether there is alternative service (i.e. replaceable) associatedwith the aborted CMT. If there is an alternative, then it selects the appropriate alternative service among the rankedservices (using the selection process). The coordinator then communicates with the respective CSC regarding thecommit or abort of the alternative CMT (repeating steps 4–7). If the aborted CMT does not have an alternative,then it takes one of two actions. If the CMT is defined as vital (by the user), then the coordinator records the abortof the MST and sends abort messages to the CSCs (if they have voted to commit). However, if the aborted CMTdoes not have an alternative transaction and is non-vital, then the coordinator ignores such aborts as the MST can

still be committed if the desired number of other CMTs are committed.



Algorithm 2: Component Service Coordinator

1. receive request for a new CMTi ∈ MST2. record start(CMTi) in a log-file3. process CMTi

4. check commit/abort of CMTi

if CMTi = commit {record commit (CMTi)send commit-votei to coordinatorwait for final-decision (from coordinator)

}else if CMTi = abort {

record abort (CMTi)send abort-votei to coordinator

}5. receive final-decision

if final-decision = commitrecord final-commit (CMTi)record termination of CMTi

else if final-decision = abortcompensate CMTi

record compensation of CMTi

6. record termination of CMTi

Algorithm 2. The CSC receives a request for a new CMT (step 1). It records the start of the CMT in a log file(step 2). After processing the CMT (step 3) the CSC checks whether a CMT is committed or aborted (step 4).If the CMT is committed, the CSC records the commit of the CMT and sends a commit vote to the coordinator.It then waits for the final decision regarding the commit of the CMT from the coordinator, i.e. whether the CMTwill be committed or compensated. This ultimately depends on the commit of other CMTs. However, if the CMTis aborted, then the CSC records the abort of the CMT and sends an abort vote to the coordinator. It then marksthe termination of the CMT. When the CSC receives the final decision from the coordinator (step 5) it acts asfollows. If the final decision is to commit, then it records the commit of the CMT and marks the termination ofthe CMT. However, if the final decision is to abort, then it executes the compensating transaction to compensatefor the effects of the committed CMT in order to ensure the data consistency of the underlying data source.

After compensation, it marks the termination of the CMT (step 6).

5. EVALUATION

This section describes the evaluation of the proposed approach. The evaluation is performed in twostages. First, we evaluate the proposed selection process for selecting mobile services. Based on theselection process, we then evaluate the throughput (i.e. commit rate) of the proposed approach.

5.1. Mobile service selection

We evaluate the selection process using a case study of a hotel booking service. We use three hotelservices, which are represented as HS1, HS2, and HS3. The selection process takes into account five



QoS attributes (or criteria) in order to evaluate the hotel services. Note that the QCMA is not limitedto these attributes. It is capable of accommodating other QoS attributes.

These attributes include service reliability (a1), usability (a2), price/cost (a3), availability (a4), andefficiency (a5). A service is considered to be reliable if it does not encounter failures for a specifictime interval. Usability is the ease or simplicity of use of a service as experienced by a mobile user.The service price is the cost users have to pay to use the service per transaction. The availability of aservice is defined as its availability for a specific time interval. The efficiency is defined as the responsetime of a service.

The procedure for evaluating service QoS is described as follows.Each mobile user (Userk) has to evaluate the above list of attributes, SQoS = {a1, a2, a3, a4, a5},

and then assign an ordering preference to the alternatives HS1, HS2, and HS3. User preferences arerepresented by a ‘preference relation’ [25] which can be obtained directly or can be transformed froma ‘preference ordering’ [28]. For example, pk

ij characterizes the preference degree between criteria ai

and aj expressed by Userk . The value of pcij is the collective preference which is an aggregation of k

users’ preferences, {p1ij, . . . , pk

ij}, by means of a fuzzy majority [29]. The final collective preference,pc

ij, of all user preferences in this experiment is given by the following matrix:

pcij =

0.5000 0.6500 0.3500 0.3625 0.51250.3500 0.5000 0.2000 0.2125 0.36250.6500 0.8000 0.5000 0.5125 0.66250.6375 0.7875 0.4875 0.5000 0.65000.4875 0.6375 0.3375 0.3500 0.5000

As described above, two fuzzy ranking methods are used to identify the priorities of attributes fromcollective preference, pc

ij: the quantifier guided non-dominance degree (QGNDD) and the quantifierguided dominance degree (QGDD) [30]. QGDD can quantify the ordering preference dominance thatai has over all of the other criteria using the fuzzy majority concept and QGNDD specifies the degree towhich ai is not dominated by a fuzzy majority of the remaining criteria [28]. The aim of the proceduresis to identify the importance of these criteria and to calculate their weightings according to the users’preferences.

Figure 6 shows the result of the importance of QoS attributes obtained from QGDD and QGNDD.The value of wai (i = 1 . . . 5) represents the weighting of attribute ai . The importance of the criteriacan be concluded as

a3 (service price) > a4 (availability) > a1 (reliability) > a5 (efficiency) > a2 (usability)

The equation for evaluating QoS attributes with different weightings can be represented as

QoS = 0.2625a3 + 0.2562a4 + 0.1875a1 + 0.1813a5 + 0.1125a2

After the derived QoS formula has been obtained, the evaluation process of the hotel services has tobe carried out again by inviting a group of users to assess these weightings. The average score of anumber of users’ feedback is shown in Table I.

The consensus weights for attributes from the QGDD are then multiplied by the average score toobtain the final score for three alternatives:

QoS score = [0.2226 (HS1) 0.2281 (HS2) 0.2004 (HS3)]Thus, the complete preference order of the alternative services is HS2 > HS1 > HS3. The MST firstuses HS2 to process user’s requests. If HS2 fails it will then use HS1 and if HS1 fails it then uses HS3.



0.1875

0.1125

0.2625 0.2562

0.1813

0.23380.2353

0.1323

0.2015 0.1971

0.1

0.12

0.14

0.16

0.18

0.2

0.22

0.24

0.26

0.28

wa1 wa2 wa3 wa4 wa5

CW from QGDD CW from QGNDD

Figure 6. The QoS attributes obtained from the consensus weights from QGDD and QGNDD.

Table I. Weightings of services according to QoS attributes.

QoS attributes

Hotel service a3 a4 a1 a5 a2

HS1 0.4578 0.2648 0.1278 0.0412 0.0284HS2 0.3830 0.2642 0.1648 0.1140 0.0740HS3 0.1428 0.1800 0.4010 0.1536 0.1226

5.2. MST throughput

Throughput is defined as the number of MSTs that are committed (successfully completed).We simulate simple MSTs that book hotel rooms and accordingly make payment for the rooms, i.e. aMST comprises one CMT for hotel booking (and two alternatives in the case of failures) and anotherCMT for making payment for the hotel room. The MST first uses the most preferred mobile service,HS2, for hotel booking. However, if it fails then its alternative service is executed to book a room atthe next preferred hotel (i.e. HS1). If this alternative also fails then the next alternative is executed tobook a room at the third hotel (i.e. HS3). In the experiments, we assume that the money transfer CMTdoes not encounter failures. However, should such a failure occur then the transaction will be aborted.

The commit of a MST is simulated as a commit probability, which represents a probability that aMST will successfully complete. Failure is simulated as an abort probability. We conducted a rangeof experiments with varying degrees of failure, and these failures were simulated using differentprobabilities with probability values ranging from 0 (no failures) to 1 (complete failures). The outputsof these experiments are shown in Figures 7 and 8. Figure 7 represents the commit scenario of aMST. It shows that our approach increases the throughput (commit) of a MST when compared with



0

0.2

0.4

0.6

0.8

1

1.2

1 2 3 4 5 6 7 8 9 10 11

Case

Com

mit

No Alternative

1 Alternative

2 Alternatives

Figure 7. Commit scenario of a MST.

0

0.2

0.4

0.6

0.8

1

1.2

1 2 3 4 5 6 7 8 9 10 11

Case

Abort

No Alternative

1 Alternative

2 Alternatives

Figure 8. Abort scenario of a MST.



other approaches that do not consider alternative transactions. Figure 8 represents the abort scenarioof a MST. Again this figure demonstrates that our approach significantly reduces the abort chances ofa MST.

6. CONCLUSION

In this paper we have presented a new approach for MSTs. Unlike existing approaches, we haveconsidered classical as well as mobile service-specific QoS attributes. Considering a limited number ofclassical QoS attributes is inappropriate given the availability of a large number of alternative mobileservices. Such QoS attributes are important, specifically from the mobile user’s perspective, as theydemand services that fulfill their QoS needs. The development of a group consensus approach reducedsubjective opinions given by mobile users in the selection of alternatives.

Our approach provides users with services that meet their QoS demands and also increases thethroughput (i.e. the number of committed transactions) of their transactions in the case of failures.The experiments show the average successful rate with one and two alternatives is higher than the casewith no alternative in the commit scenario of a MST. Accordingly, the experiments show that the abortrate is also minimized. However, the use of alternative service transactions increases the throughput asthey affect the overall performance of a MST. However, evaluation of the overall performance is out ofthe scope of this paper and will be the subject of future research work.

ACKNOWLEDGEMENT

This research is partially supported by grant No. NSC 95-2416-H-168-007 from the National Science Council ofthe Republic of China.

REFERENCES

1. Pilioura T, Tsalgatidou A, Hadjiefthimiades S. Scenarios of using Web services in m-commerce. ACM SIGecom Exchanges2003; 3(4):28–36.

2. Maamar Z, Sheng Q, Benatallah B. Selection of Web services for composition using location of provider hosts criterion.Proceedings of the 2003 Ubiquitous Mobile Information and Collaboration Systems Workshop (UMICS 2003), Austria,June 2003.

3. Maamar Z, Sheng QZ, Benatallah B. On composite Web services provisioning in an environment of fixed and mobilecomputing resources. Information Technology and Management 2004; 5(3–4):251–270.

4. Elmagarmid AK. Database Transaction Models for Advanced Applications. Morgan Kaufman: San Francisco, CA, 1992.5. Kuramitsu K, Sakamura K. Towards ubiquitous database in mobile commerce. Proceedings of the 2nd ACM International

Workshop on Data Engineering for Wireless and Mobile Access, CA, May 2001. ACM Press: New York, 2001.6. Younas M, Eaglestone B, Chao K. A low latency resilient protocol for e-business transactions. International Journal of

Web Engineering and Technology 2004; 1(3):278–296.7. Wang P, Chao K-M, Lo C-C, Huang C-L, Younas M. A fuzzy outranking approach in risk analysis of Web service security.

Cluster Computing: The Journal of Networks, Software Tools and Applications, to appear.8. Younas M, Awan I, Chao K. Network-centric strategy for mobile transactions. Journal of Interconnection Networks 2004;

5(3):329–350.9. Huang C, Lo C, Chao K, Younas M. A moderated fuzzy Web services discovery method. Journal of Information and

Software Technology 2006; 48(6):410–423.10. Kumar V, Prabhu N, Dunham M, Seydim YA. TCOT—a timeout-based mobile transaction commitment protocol. IEEE

Transactions on Computers 2002; 5(1):1212–1218.



11. Alvarado P, Rouvoy R, Merle P. Self adaptive component based transaction commit management. Proceedings of the 4thInternational Workshop on Adaptive and Reflective Middleware (ARM), Grenoble, France, November 2005. ACM Press:New York, 2005.

12. Varshney U, Ravikumar K. Transaction and networking support for group oriented mobile commerce services. Proceedingsof the 1st Conference on Broadband Wireless Networks (BroadWISE), San Jose, CA, October 2004.

13. Lee M, Helal S. HiCoMo: High commit mobile transactions. Distributed and Parallel Databases 2002; 11(1):73–92.14. Dunham MH, Helal A, Balakrishnan S. A mobile transaction model that captures both the data and movement behavior.

ACM/Baltzer Journal on Special Topics in Mobile Networks and Applications 1997; 2(2):149–162.15. Dirckze RA, Gruenwald L. A pre-serialization transaction management technique for mobile multidatabases. Mobile

Networks and Applications 2000; 5(4):111–321.16. Madria S, Bhargava B. A transaction model to improve data availability in mobile computing. Distributed and Parallel

Databases 2001; 10(2):127–160.17. Dunham M, Kumar V. Impact of mobility on transaction management. Proceedings of the ACM International Workshop

on Data Engineering for Wireless and Mobile Access, Seattle, WA, 1999. ACM Press: New York, 1999.18. Menasce D. Composing Web services: A QoS view. IEEE Internet Computing 2004; 8(6):88–90.19. Huang C-L. An approach to consensus-based service discovery. PhD Dissertation, National Chiao Tung University,

Taiwan, September 2006.20. Younas M, Eaglestone M, Holton R. A formal treatment of a SACReD protocol for multidatabase Web transactions.

Proceedings of the 11th International Conference on Database and Expert Applications, London, 2000 (Lecture Notesin Computer Science, vol. 1873). Springer: Berlin, 2000.

21. Huang C, Chao K, Lo C, Wang P, Chung J. Service discovery through consensus-based preferences. International Journalof Computer Systems Science and Engineering 2006; 21(4):255–262.

22. Zimmermann H. Fuzzy Set Theory and its Applications. Kluwer: Dordrecht, 1991.23. Yen J, Langari R. Fuzzy Logic: Intelligence, Control, and Information. Prentice-Hall: Englewood Cliffs, NJ, 1999.24. Hsu H, Chen C. Aggregation of fuzzy opinions under group decision making. Fuzzy Sets and Systems 1996; 79:279–285.25. Chiclana F, Herrera F, Herrera-Viedma E. Integrating multiplicative preference relations in a multipurpose decision-making

model based on fuzzy preference relations. Fuzzy Sets and Systems 2001; 122(2):277–291.26. Yager R. On ordered weighted averaging aggregation operators in multi criteria decision making. IEEE Transactions on

Systems, Man, and Cybernetics 1988; 18:183–190.27. Smith M, Welty C, McGuinness D. OWL Web Ontology Language Guide. W3C Recommendation, February 2004.

Available at: http://www.w3.org/TR/2004/REC-owl-guide-20040210/.28. Chiclana F, Herrera F, Herrera-Viedma E. Integrating three representation models in fuzzy multipurpose decision making

based on fuzzy preference relations. Fuzzy Sets and Systems 1998; 97:33–48.29. Chiclana F, Herrera F, Herrera-Viedma E. A classification method of alternatives for multiple preference ordering criteria

based on fuzzy majority. Journal of Fuzzy Mathematics 1996; 34:224–229.30. Herrera-Viedma E, Herrera F, Chiclana F. A consensus model for multiperson decision making with different preference

structures. IEEE Transactions on Systems, Man, and Cybernetics 2002; 32:394–402.


Documents

QoS-aware mobile service transactions in a wireless environment